Forecasting US Inflation with a Dynamic Factor Model

Abstract:

The paper studies the pseudo real-time forecasting performance of three different factor models. We compare the method recently proposed by Forni et al. (2015) and Forni et al. (2014) with those proposed in Forni et al. (2005) and Stock and Watson (2002a) within a real data forecasting exercise. A large panel of macroeconomic and financial time series for the US economy which includes the Great Recession and the subsequent recovery is employed. In a rolling window framework, we find that the first two methods, based on spectral estimation, outperform the third. Substantial gains from regularized combinations of different inflation forecasts produced with the model in Forni et al. (2015) are also found.